cxas-sim-eval

Pass

Audited by Gen Agent Trust Hub on May 15, 2026

Risk Level: SAFE
Full Analysis
  • [PROMPT_INJECTION]: The skill implements an indirect prompt injection surface in 'scripts/convert_eval.py' and 'scripts/run_evals.py'. It interpolates external data, such as CXAS golden evaluations, agent logs, and internal monologues, directly into Gemini prompts without using strict boundary markers or sanitization. This is a design-inherent characteristic for a tool intended to analyze and convert AI-generated content.
  • [COMMAND_EXECUTION]: The SKILL.md documentation includes standard environment setup instructions using shell commands, such as 'gcloud auth login' and checking for the 'cxas_scrapi' package. These commands are routine for developers working within the Google Cloud ecosystem.
  • [EXTERNAL_DOWNLOADS]: The skill identifies a dependency on 'cxas_scrapi', which is a vendor-specific Python package provided by the author for interfacing with the SCRAPI framework. This is a trusted resource consistent with the skill's stated purpose.
Audit Metadata
Risk Level
SAFE
Analyzed
May 15, 2026, 07:56 PM
Security Audit — agent-trust-hub — cxas-sim-eval